Describe the issue:

Followup of#27957 because the issue is not fixed fornumpy.linalg.inv.

numpy.linalg.inv apparently loses some typing information.

The following snippet is a test case demonstrating the issue. In my real code, I tried to pass inv_test to a function that only acceptsNDArray[np.float64]. I'd expect an inversion to keep the same float size in type.

Reproduce the code example:

importnumpyasnpimportnumpy.typingasnpttest:npt.NDArray[np.float64]=np.array([1.0,2.0,3.0],dtype=np.float64)inv_test:npt.NDArray[np.float64]=np.linalg.inv(test)

Error message:

test.py:5: error: Incompatible typesin assignment (expression hastype"ndarray[tuple[Any, ...], dtype[floating[Any]]]", variable hastype"ndarray[tuple[Any, ...], dtype[float64]]")  [assignment]Found 1 errorin 1 file (checked 1source file)

Python and NumPy Versions:

numpy: 2.3.5
python: 3.14.0 (main, Oct 7 2025, 09:34:52) [GCC 14.3.0]

Type-checker version and settings:

❯ mypy --versionmypy 1.19.0 (compiled: yes)

I usedmypy test.py andmypy --local-partial-types --allow-redefinition-new test.py, same result.

Additional typing packages.

typing_extensions: 4.15.0